AI-powered talent search CLI for natural language queries, exposable as an MCP server over stdio transport
Three tools are explicitly registered in src/mcp/server.ts with basic schemas and descriptions. Naming follows verb_noun conventions (talent_search, talent_detail, talent_refine), which is good. However, descriptions are minimal (50-80 chars) and lack the depth required by the rubric's baseline of 194 chars average. Parameter descriptions exist but are generic. Most critically, output schemas are completely undocumented, the tools return JSON-stringified content with no explanation of what fields agents should expect. Error handling is absent; no guidance on retryability, invalid inputs, or recovery paths. Security is reasonable (read-only tools, no credentials in params) but not explicitly documented. The server structure is clean, but definition quality falls short of production baseline.
Get detailed profile information for a candidate at a given index from the last search in a session.
Refine an existing search with additional criteria. Continues the conversation in the given session.
Search for talent profiles using natural language. Returns a list of matching profiles with summaries.
Output schemas not documented. All three tools return JSON-stringified text with no documented schema of what fields exist. Agents cannot infer what data is available or plan downstream chaining.
Descriptions are minimal (50 - 80 chars) and lack context. Baseline is 194 chars. talent_search describes the action but omits when to use it vs talent_refine. talent_detail lacks explanation of the index parameter source (from search results).
No error handling guidance. All three tools could fail (invalid session, out-of-bounds index, API errors) but provide no recovery hints or categorization (retryable vs user-fixable). Agents have no recovery path.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 46 | - | v1 |
Parameter description 'Session ID to continue a previous search conversation' assumes agents understand session mechanics. No guidance on how to obtain a session, whether it persists, or what happens if it expires.
talent_detail index parameter lacks bounds guidance. Description says 'Zero-based index' with minimum 0, but no maximum. What happens if index exceeds result count? Agents need explicit error behavior.
No output field naming consistency documented. If talent_search returns {'candidates': [...], 'total': N}, then talent_detail must reference fields agents can extract. Currently, all three tools stringify JSON without specifying field contracts.